DOCSLIB.ORG

Intelligent User Interfaces for Music Discovery

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Intelligent User Interfaces for Music Discovery Knees, P., et al. (2020). Intelligent User Interfaces for Music Discovery. Transactions of the International Society for Music Information 7,60,5 Retrieval, 3(1), pp. 165–179. DOI: https://doi.org/10.5334/tismir.60 OVERVIEW ARTICLE Intelligent User Interfaces for Music Discovery Peter Knees*, Markus Schedl† and Masataka Goto‡ Assisting the user in finding music is one of the original motivations that led to the establishment of Music Information Retrieval (MIR) as a research field. This encompasses classic Information Retrieval inspired access to music repositories that aims at meeting an information need of an expert user. Beyond this, however, music as a cultural art form is also connected to an entertainment need of potential listeners, requiring more intuitive and engaging means for music discovery. A central aspect in this process is the user interface. In this article, we reflect on the evolution of MIR-driven intelligent user interfaces for music browsing and discovery over the past two decades. We argue that three major developments have transformed and shaped user interfaces during this period, each connected to a phase of new listening practices. Phase 1 has seen the development of content-based music retrieval interfaces built upon audio processing and content description algorithms facilitating the automatic organization of repositories and finding music according to sound qualities. These interfaces are primarily connected to personal music collections or (still) small commercial catalogs. Phase 2 comprises interfaces incorporating collaborative and automatic semantic description of music, exploiting knowledge captured in user-generated metadata. These interfaces are connected to collective web platforms. Phase 3 is dominated by recommender systems built upon the collection of online music interaction traces on a large scale. These interfaces are connected to streaming services. We review and contextualize work from all three phases and extrapolate current developments to outline possible scenarios of music recommendation and listening interfaces of the future. Keywords: music access; music browsing; user interfaces; content-based MIR; community metadata; recommender systems 1. Introduction digital repositories of acoustic content, retrieving relevant With its origins in Information Retrieval research, a pieces from music collections, and discovering new fundamental goal of Music Information Retrieval (MIR) as music, interfaces based on querying, visual browsing, a dedicated research field in the year 2000 was to develop or recommendation have facilitated new modes of technology to assist the user in finding music, information interaction. about music, or information in music (Byrd and Fingerhut, Revisiting an early definition of MIR by Downie (2004) 2002). Since then, also driven by the developments as “a multidisciplinary research endeavor that strives to in content-based analysis, semantic annotation, and develop innovative content-based searching schemes, personalization, intelligent music applications have had novel interfaces, and evolving networked delivery significant impact on people’s interaction with music. mechanisms in an effort to make the world’s vast store These applications comprise “active music-listening of music accessible to all” 16 years later, reveals that interfaces” (Goto and Dannenberg, 2019) which augment these developments were in fact intended and implanted the process of music listening to increase engagement of into MIR from the beginning. Given the music industry the user and/or give deeper insights into musical aspects landscape and how people listen to music today,1 this also to musically less experienced listeners. For accessing visionary definition has undoubtedly stood the test of time. In this paper, we reflect on the evolution of MIR-driven user interfaces for music browsing and discovery over * Faculty of Informatics, TU Wien, Vienna, AT the past two decades—from organizing personal music † Institute of Computational Perception, Johannes Kepler University Linz, AT collections to streaming a personalized selection from ‡ National Institute of Advanced Industrial Science and “the world’s vast store of music”. Therefore, we connect Technology (AIST), JP major developments that have transformed and shaped Corresponding author: Peter Knees ([email protected]) MIR research in general, and user interfaces in particular, 166 Knees et al: Intelligent User Interfaces for Music Discovery to prevalent and emerging listening practices at the time. with the title list and mere text searches based on We identify three main phases that have each laid the bibliographic information are useful enough to browse foundation for the next and review work that focuses on the whole collection to choose pieces to listen to. However, the specific aspects of these phases. as the accessible collection grows and becomes largely First, we investigate the phase of growing digital unfamiliar, such simple interfaces become insufficient personal music collections and interfaces built upon (Cunningham et al., 2017; Cunningham, 2019), and new intelligent audio processing and content description research approaches targeting the retrieval, classification, algorithms in Section 2. These algorithms facilitate the and organization of music emerge. automatic organization of repositories and finding music “Intelligent” interfaces for music retrieval become in personal collections, as well as commercial repositories, a research field of interest with the developments in according to sound qualities. Second, in Section 3, we content-based music retrieval (Casey et al., 2008). A investigate the emergence of collective web platforms and landmark in this regard is the development of query by their exploitation for listening interfaces. The extracted humming systems (Kageyama et al., 1993) and search user-generated metadata often pertains to semantic engines indexing sound properties of loudness, pitch, and descriptions and complements the content-based methods timbre (Wold et al., 1996) that initiate the emancipation of that facilitated the developments of the preceding phase. music search systems from traditional text- and metadata- This phase also constitutes an intermediate step towards based indexing and query interfaces. While interfaces are exploitation of collective listening data, which is the still very much targeted at presenting results in sequential driving force behind the third, and ongoing phase, which order according to relevance to a query, starting in the is connected to streaming services (Section 4). Here, the early 2000s, MIR research proposes several alternatives to collection of online music interaction traces on a large facilitate music discovery. scale and their exploitation in recommender systems are defining elements. Extrapolating these and other ongoing 2.1 Map-based music browsing and discovery developments, we outline possible scenarios of music Interfaces that allow content-based searches for music recommendation and listening interfaces of the future in retrieval are useful when people can formulate good Section 5. queries and especially when users are looking for a Note that the phases we identify in the evolution of user particular work, but sometimes it is difficult to come interfaces for music discovery also reflect the “three ages up with an appropriate query when faced with a huge of MIR” as described by Herrera (2018). Herrera refers to music collection and vague search criteria. Interfaces for these three phases as “the age of feature extractors”, “the music browsing and discovery are therefore proposed to age of semantic descriptors” and “the age of context-aware let users encounter unexpected but interesting musical systems”, respectively. We further agree on the already pieces or artists. Visualization of a music collection is ongoing “age of creative systems” that builds upon MIR one way to provide users with various bird’s-eye views to facilitate new interfaces that support creativity as we and comprehensive interactions. The most popular discuss in Section 5. We believe that this strong alignment visualization is to project musical pieces or artists onto gives further evidence of the pivotal role of intelligent a 2D or 3D space (“map”) by using music similarity. 2D user interfaces in the development of MIR. While user visualizations also lend themselves to being applied on interfaces, especially in the early phases, were often tabletop interfaces for intuitive access and interaction (e.g. mere research prototypes, their development is tightly Julià and Jordà, 2009). The trend of spatially arranging intertwined with ongoing trends. Thus, they provide collections for exploration can be seen throughout the essential gauges to the state of the art, and, even beyond, past 20 years and is still unbroken, cf. Figure 1. give perspective of what could be possible. One of the earliest interfaces is GenreSpace by Tzanetakis et al. (2001) that visualizes musical pieces with 2. Phase 1: Content-Based Music Retrieval genre-specific colors in a 3D space (see Figure 1(a) for Interfaces a greyscale image). Coloring of each piece is determined The late 1990s see two pivotal developments. On one by automatic genre classification. The layout of pieces is hand, the Internet gets established as mainstream determined by principal component analysis (PCA), which communication medium and distribution channel. On projects high-dimensional audio feature vectors into 3D the other hand, technological
Load more

Recommended publications
  • Movie Genome: Alleviating New Item Cold Start in Movie Recommendation
    User Modeling and User-Adapted Interaction https://doi.org/10.1007/s11257-019-09221-y Movie genome: alleviating new item cold start in movie recommendation Yashar Deldjoo, et al. [full author details at the end of the article] Received: 23 December 2017 / Accepted in revised form: 19 January 2019 © The Author(s) 2019 Abstract As of today, most movie recommendation services base their recommendations on col- laborative filtering (CF) and/or content-based filtering (CBF) models that use metadata (e.g., genre or cast). In most video-on-demand and streaming services, however, new movies and TV series are continuously added. CF models are unable to make predic- tions in such a scenario, since the newly added videos lack interactions—a problem technically known as new item cold start (CS). Currently, the most common approach to this problem is to switch to a purely CBF method, usually by exploiting textual meta- data. This approach is known to have lower accuracy than CF because it ignores useful collaborative information and relies on human-generated textual metadata, which are expensive to collect and often prone to errors. User-generated content, such as tags, can also be rare or absent in CS situations. In this paper, we introduce a new movie recommender system that addresses the new item problem in the movie domain by (i) integrating state-of-the-art audio and visual descriptors, which can be automati- cally extracted from video content and constitute what we call the movie genome; (ii) exploiting an effective data fusion method named canonical correlation analysis, which was successfully tested in our previous works Deldjoo et al.
    [Show full text]
  • Recommender Systems User-Facing Decision Support Systems
    Recommender Systems User-Facing Decision Support Systems Michael Hahsler Intelligent Data Analysis Lab (IDA@SMU) CSE, Lyle School of Engineering Southern Methodist University EMIS 5/7357: Decision Support Systems February 22, 2012 Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 1 / 44 Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 2 / 44 Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 3 / 44 Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 4 / 44 Table of Contents 1 Recommender Systems & DSS 2 Content-based Approach 3 Collaborative Filtering (CF) Memory-based CF Model-based CF 4 Strategies for the Cold Start Problem 5 Open-Source Implementations 6 Example: recommenderlab for R 7 Empirical Evidence Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 5 / 44 Decision Support Systems ? Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 6 / 44 Decision Support Systems \Decision Support Systems are defined broadly [...] as interactive computer-based systems that help people use computer communications, data, documents, knowledge, and models to solve problems and make decisions." Power (2002) Main Components: Knowledge base Model User interface Michael Hahsler (IDA@SMU) Recommender Systems EMIS 5/7357: DSS 6 / 44 ! A recommender system is a decision support systems which help a seller to choose suitable items to offer given a limited information channel. Recommender Systems Recommender systems apply statistical and knowledge discovery techniques to the problem of making product recommendations. Sarwar et al. (2000) Advantages of recommender systems (Schafer et al., 2001): Improve conversion rate: Help customers find a product she/he wants to buy.
    [Show full text]
  • Pandora's Music Recommender 1 Introduction to Pandora 2 Description of Pandora Problem Space
    Pandora’s Music Recommender Michael Howe 1 Introduction to Pandora One of the great promises of the internet and Web 2.0 is the opportunity to expose people to new types of content. Companies like Amazon and Netflix provide customers with ideas for new items to purchase based on current or previous selections. For instance, someone who rented “Star Wars” at Netflix might be presented with “The Matrix” as another movie to rent. The challenge in this strategy is to make suggestions in a reasonable amount of time that the user will mostly like based on the known list of what the user already enjoys. People will only pay attention to the recommendations of a service a finite number of times before they lose trust. Repeatedly suggest content that the user hates and the user will look for new content elsewhere. Also, a user will only pay attention to a service’s recommendation if they arrive in a reasonable amount of time. Make the user wait longer than they are willing and they will again turn elsewhere for content suggestions. Accuracy and speed are critical to a service’s success. Pandora exposes people to music with an online radio station where a user builds up “stations” based on musical interests. The user indicates in each station one or more songs or artists that he or she likes. Based on these preferences Pandora plays similar songs that the user might also like. As Pandora plays the user can further refine the station by giving a “thumbs up” or a “thumbs down” to a particular song.
    [Show full text]
  • Music Recommendation Algorithms: Discovering Weekly Or Discovering
    MUSIC RECOMMENDATION ALGORITHMS Music Recommendation Algorithms: Discovering Weekly or Discovering Weakly? Jennie Silber Muhlenberg College, Allentown, PA, USA Media & Communication Department Undergraduate High Honors Thesis April 1, 2019 1 MUSIC RECOMMENDATION ALGORITHMS 2 MUSIC RECOMMENDATION ALGORITHMS Abstract This thesis analyzes and assesses the cultural impact and economic viability that the top music streaming platforms have on the consumption and discovery of music, with a specific focus on recommendation algorithms. Through the support of scholarly and journalistic research as well as my own user experience, I evaluate the known constructs that fuel algorithmic recommendations, but also make educated inferences about the variables concealed from public knowledge. One of the most significant variables delineated throughout this thesis is the power held by human curators and the way they interact with algorithms to frame and legitimize content. Additionally, I execute my own experiment by creating new user profiles on the two streaming platforms popularly used for the purpose of discovery, Spotify and SoundCloud, and record each step of the music discovery process experienced by a new user. After listening to an equal representation of all genre categories within each platform, I then compare the genre, release year, artist status, and content promotion gathered from my listening history to the algorithmically-generated songs listed in my ‘Discover Weekly’ and ‘SoundCloud Weekly’ personalized playlists. The results from this experiment demonstrate that the recommendation algorithms that power these discovery playlists intrinsically facilitate the perpetuation of a star- driven, “winner-take-all” marketplace, where new, popular, trendy, music is favored, despite how diverse of a selection the music being listened to is.
    [Show full text]
  • UNIVERSITY of CALIFORNIA, SAN DIEGO Bag-Of-Concepts As a Movie
    UNIVERSITY OF CALIFORNIA, SAN DIEGO Bag-of-Concepts as a Movie Genome and Representation A Thesis submitted in partial satisfaction of the requirements for the degree Master of Science in Computer Science by Colin Zhou Committee in charge: Professor Shlomo Dubnov, Chair Professor Sanjoy Dasgupta Professor Lawrence Saul 2016 The Thesis of Colin Zhou is approved, and it is acceptable in quality and form for publication on microfilm and electronically: Chair University of California, San Diego 2016 iii DEDICATION For Mom and Dad iv TABLE OF CONTENTS Signature Page........................................................................................................... iii Dedication.................................................................................................................. iv Table of Contents....................................................................................................... v List of Figures and Tables.......................................................................................... vi Abstract of the Thesis................................................................................................ vii Introduction................................................................................................................ 1 Chapter 1 Background............................................................................................... 3 1.1 Vector Space Models............................................................................... 3 1.1.1. Bag-of-Words........................................................................
    [Show full text]
  • A Movie Recommender System: MOVREC Using Machine Learning Techniques Ashrita Kashyap1, Sunita
    ISSN 2321 3361 © 2020 IJESC Research Article Volume 10 Issue No.6 A Movie Recommender System: MOVREC using Machine Learning Techniques Ashrita Kashyap1, Sunita. B2, Sneh Srivastava3, Aishwarya. PH4, Anup Jung Shah5 Department of Computer Science & Engineering SAIT, Bengaluru, Karnataka, India Abstract: This paper discuss about recommendations of the movies. A movie recommendation is important in our social life due to its strength in providing enhanced entertainment. Such a system can suggest a set of movies to users based on their interest, or the popularities of the movies. A recommendation system is used for the purpose of suggesting items to purchase or to see. They direct users towards those items which can meet their needs through cutting down large database of Information. A recommender system, or a recommendation system (sometimes replacing 'system' with a synonym such as platform or engine), is a subclass of information filtering system that seeks to predict the "rating" or "preference" a user would give to an item[1][2]. They are primarily used in commercial applications. MOVREC also help users to find the movies of their choices based on the movie experience of other users in efficient and effective manner without wasting much time in useless browsing. Keywords: Filtering, Recommendation System, Recommender. I. INTRODUCTION emphasizing other attributes when a user "likes" a song. This is an example of a content-based approach. Recommender systems usually make use of either or both collaborative filtering and content-based filtering (also known as Each type of system has its strengths and weaknesses. In the the personality-based approach),[3] as well as other systems above example, Last.fm requires a large amount of information such as knowledge-based systems.
    [Show full text]
  • Five Approaches to Collecting Tags for Music
    FIVE APPROACHES TO COLLECTING TAGS FOR MUSIC Douglas Turnbull Luke Barrington Gert Lanckriet UC San Diego UC San Diego UC San Diego [email protected] [email protected] [email protected] ABSTRACT There are a number of key concepts to consider when com- paring these approaches. The cold start problem refers to the We compare five approaches to collecting tags for music: fact songs that are not annotated cannot be retrieved. This conducting a survey, harvesting social tags, deploying anno- problem is related to popularity bias in that popular songs (in tation games, mining web documents, and autotagging audio the short-head) tend to be annotated more thoroughly than content. The comparison includes a discussion of both scala- unpopular songs (in the long-tail)[11]. This often leads bility (financial cost, human involvement, and computational to a situation in which a short-head song is ranked above resources) and quality (the cold start problem & popularity a long-tail song despite the fact that the long-tail song may bias, strong vs. weak labeling, vocabulary structure & size, be more semantically relevant. We prefer an approach that and annotation accuracy). We then describe one state-of- avoids the cold start problem (e.g., autotagging). If this is the-art system for each approach. The performance of each not possible, we prefer approaches in which we can explic- system is evaluated using a tag-based music information itly control which songs are annotated (e.g., survey, games), retrieval task. Using this task, we are able to quantify the rather than an approach in which only the more popular songs effect of popularity bias on each approach by making use are annotated (e.g., social tags, web documents).
    [Show full text]
  • DEEP LEARNING for MUSIC RECOMMENDATION: Machine Listening & Collaborative Filtering
    DEEP LEARNING FOR MUSIC RECOMMENDATION: Machine Listening & Collaborative Filtering ORIOL NIETO [email protected] SEMINAR ON MUSIC KNOWLEDGE EXTRACTION USING MACHINE LEARNING POMPEU FABRA UNIVERSITY BARCELONA DECEMBER 4, 2016 Outline • Pandora Overview • Large Scale Music Recommendation with Deep Learning: • Machine Listening • Collaborative Filtering Outline • Pandora Overview • Large Scale Music Recommendation with Deep Learning: • Machine Listening • Collaborative Filtering TODAY 78 M 22 H 75 B + Monthly Listeners Average Listening per Month Thumbs 10 B 180 K 55 % Stations Artists Spinning per Month of all Music Streaming in USA Content&Based Collective) Intelligence Personalized Filtering Outline • Pandora Overview • Large Scale Music Recommendation with Deep Learning: • Machine Listening • Collaborative Filtering Content&Based Collective) Intelligence Personalized Filtering Short Time Fourier Transform MY GUITAR GENTLY WEEPS - THE BEATLES Extracting Information Beats Downbeats Chords Structure “electric guitar”, “drums”, “male singer”, “pop”, “classic rock” Hand Crafted Features Beats Downbeats Chords MFCCs Structure Spectral Rolloff Spectral Centroid Structural Features Chromagrams Tempograms “electric guitar”, CERNs “drums”, … “male singer”, “pop”, “classic rock” Learned Features Beats Downbeats Chords Structure DEEP LEARNING (Mostly Convolutional Neural Networks) “electric guitar”, “drums”, “male singer”, “pop”, http://blogs-images.forbes.com/kevinmurnane/files/2016/03/google-deepmind-artificial-intelligence-2-970x0-970x646.jpg
    [Show full text]
  • A Hybrid Approach to Music Recommendation: Exploiting Collaborative Music Tags and Acoustic Features Jaime C
    UNF Digital Commons UNF Graduate Theses and Dissertations Student Scholarship 2014 A Hybrid Approach to Music Recommendation: Exploiting Collaborative Music Tags and Acoustic Features Jaime C. Kaufman University of North Florida Suggested Citation Kaufman, Jaime C., "A Hybrid Approach to Music Recommendation: Exploiting Collaborative Music Tags and Acoustic Features" (2014). UNF Graduate Theses and Dissertations. 540. https://digitalcommons.unf.edu/etd/540 This Master's Thesis is brought to you for free and open access by the Student Scholarship at UNF Digital Commons. It has been accepted for inclusion in UNF Graduate Theses and Dissertations by an authorized administrator of UNF Digital Commons. For more information, please contact Digital Projects. © 2014 All Rights Reserved A HYBRID APPROACH TO MUSIC RECOMMENDATION: EXPLOITING COLLABORATIVE MUSIC TAGS AND ACOUSTIC FEATURES by Jaime C. Kaufman A thesis submitted to the School of Computing in partial fulfillment of the requirements for the degree of Master of Science in Computer and Information Sciences UNIVERSITY OF NORTH FLORIDA SCHOOL OF COMPUTING December 2014 Copyright (©) 2014 by Jaime C. Kaufman All rights reserved. Reproduction in whole or in part in any form requires the prior written permission of Jaime C. Kaufman or designated representative. ii The thesis, “A Hybrid Approach to Music Recommendation: Exploiting Collaborative Music Tags and Acoustic Features,” submitted by Jaime C. Kaufman in partial fulfillment of the requirements for the degree of Master of Science in Computer and Information Sciences has been Approved by the thesis committee: Date Dr. Ching-Hua Chuan Thesis Advisor and Committee Chairperson Dr. Sherif Elfayoumy Dr. Karthikeyan Umapathy Accepted for the School of Computing: Dr.
    [Show full text]
  • Combining Content Information with an Item-Based Collaborative Filter
    Volume 2 │ Issue 2 │ 2017 Combining Content Information with an Item-Based Collaborative Filter Daryl Bagley Harding University Arkansas Eta Chapter Vol. 2(2), 2017 Article Title: Combining Content Information with an Item-Based Collaborative Filter DOI: 10.21081/AX0126 ISSN: 2381-800X Key Words: algorithms, anime, collaborative filtering, content analysis, recommendation This work is licensed under a Creative Commons Attribution 4.0 International License. Author contact information is available from the Editor at [email protected]. Aletheia—The Alpha Chi Journal of Undergraduate Scholarship • This publication is an online, peer-reviewed, interdisciplinary undergraduate journal, whose mission is to promote high quality research and scholarship among undergraduates by showcasing exemplary work. • Submissions can be in any basic or applied field of study, including the physical and life sciences, the social sciences, the humanities, education, engineering, and the arts. • Publication in Aletheia will recognize students who excel academically and foster mentor/mentee relationships between faculty and students. • In keeping with the strong tradition of student involvement in all levels of Alpha Chi, the journal will also provide a forum for students to become actively involved in the writing, peer review, and publication process. • More information and instructions for authors is available under the publications tab at www.AlphaChiHonor.org. Questions to the editor may be directed to [email protected]. Alpha Chi is a national college honor society that admits students from all academic disciplines, with membership limited to the top 10 percent of an institution’s juniors, seniors, and graduate students. Invitation to membership comes only through an institutional chapter.
    [Show full text]
  • Kandidat 109
    INFO310 0 Advanced Topics in Model-Based Information Systems Kandidat 109 Oppgaver Oppgavetype Vurdering Status Introduction Dokument Automatisk poengsum Levert Plagiarism and Declaration Dokument Automatisk poengsum Levert 1 Essay Filopplasting Manuell poengsum Levert INFO310 0 Advanced Topics in Model-Based Information Systems Emnekode INFO310 PDF opprettet 16.11.2016 16:36 Vurderingsform INFO310 Opprettet av Andreas Lothe Opdahl Starttidspunkt: 03.11.2016 14:00 Antall sider 22 Sluttidspunkt: 09.11.2016 14:00 Oppgaver inkludert Nei Sensurfrist Ikke satt Skriv ut automatisk rettede Nei 1 Kandidat 109 Seksjon 1 1 OPPGAVE Essay Upload your file here. Maximum one file. BESVARELSE Filopplasting Filnavn 9066477_cand-9333784_9157556 Filtype pdf Filstørrelse 1660.232 KB Opplastingstid 09.11.2016 09:38:21 Neste side Besvarelse vedlagt INFO310 0 Advanced Topics in Model-Based Information Systems Page 2 av 22 Kandidat 109 SEMANTICS TECHNOLOGIES IN STREAMING SERVICES By Gonzalo Molina Gallego INFO310 0 Advanced Topics in Model-Based Information Systems Page 3 av 22 Kandidat 109 INDEX INTRODUCTION ......................................................................................................................... 3 THE ARISING OF STREAMING SERVICES ............................................................................ 3 VIDEO STREAMING SERVICES .......................................................................................... 4 MUSIC STREAMING SERVICES .........................................................................................
    [Show full text]
  • Matrix Factorization and Collaborative Filtering
    10-601 Introduction to Machine Learning Machine Learning Department School of Computer Science Carnegie Mellon University Matrix Factorization and Collaborative Filtering MF Readings: (Koren et al., 2009) Matt Gormley Lecture 25 April 19, 2017 1 Reminders • Homework 8: Graphical Models – Release: Mon, Apr. 17 – Due: Mon, Apr. 24 at 11:59pm • Homework 9: Applications of ML – Release: Mon, Apr. 24 – Due: Wed, May 3 at 11:59pm 2 Outline • Recommender Systems – Content Filtering – Collaborative Filtering (CF) – CF: Neighborhood Methods – CF: Latent Factor Methods • Matrix Factorization – Background: Low-rank Factorizations – Residual matrix – Unconstrained Matrix Factorization • Optimization problem • Gradient Descent, SGD, Alternating Least Squares • User/item bias terms (matrix trick) – Singular Value Decomposition (SVD) – Non-negative Matrix Factorization • Extra: Matrix Multiplication in ML – Matrix Factorization – Linear Regression – PCA – (Autoencoders) – K-means 7 RECOMMENDER SYSTEMS 8 Recommender Systems A Common Challenge: – Assume you’re a company selling items of some sort: movies, songs, products, etc. – Company collects millions of ratings from users of their items – To maximize profit / user happiness, you want to recommend items that users are likely to want 9 Recommender Systems 10 Recommender Systems 11 Recommender Systems 12 Recommender Systems Problem Setup • 500,000 users • 20,000 movies • 100 million ratings • Goal: To obtain lower root mean squared error (RMSE) than Netflix’s existing system on 3 million held out ratings 13 Recommender Systems 14 Recommender Systems • Setup: – Items: movies, songs, products, etc. (often many thousands) – Users: watchers, listeners, purchasers, etc. (often many millions) Doctor Doctor Strange Star Trek: Beyond Zootopia – Feedback: Alice 1 5 5-star ratings, not-clicking ‘next’, purchases, etc.
    [Show full text]